Skip to content

Include dpctl4pybind11.hpp in dpnp4pybind11.hpp and switch libtensor to dpnp namespace #2862

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
antonwolfy merged 10 commits into from
Apr 20, 2026
Merged

Include dpctl4pybind11.hpp in dpnp4pybind11.hpp and switch libtensor to dpnp namespace #2862

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
ed5578f
refactor dpnp4pybind11.hpp
ndgrigorian Apr 19, 2026
f015208
remove unnecessary headers
ndgrigorian Apr 19, 2026
e5e3e4b
fix format of subtract module init file
ndgrigorian Apr 19, 2026
ffa9268
change dpctl namespace to dpnp
ndgrigorian Apr 19, 2026
4a8ef52
add back ManagedMemory and keep_args_alive
ndgrigorian Apr 20, 2026
c3a4a78
add back array include
ndgrigorian Apr 20, 2026
1d71b88
change comments referencing dpctl tensor
ndgrigorian Apr 20, 2026
e2f6d8d
address review feedback
ndgrigorian Apr 20, 2026
75da978
Merge branch 'include-dpctl-tensor' into further-refactor-dpnp4pybind11
ndgrigorian Apr 20, 2026
c4594c2
fix remaining lookup from dpctl namespace
ndgrigorian Apr 20, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
8 changes: 4 additions & 4 deletions dpnp/backend/extensions/blas/blas_py.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,15 +60,15 @@ void init_dispatch_vectors_tables(void)
blas_ns::init_syrk_dispatch_vector();
}

static dot_impl_fn_ptr_t dot_dispatch_vector[dpctl_td_ns::num_types];
static dot_impl_fn_ptr_t dotc_dispatch_vector[dpctl_td_ns::num_types];
static dot_impl_fn_ptr_t dotu_dispatch_vector[dpctl_td_ns::num_types];
static dot_impl_fn_ptr_t dot_dispatch_vector[dpnp_td_ns::num_types];
static dot_impl_fn_ptr_t dotc_dispatch_vector[dpnp_td_ns::num_types];
static dot_impl_fn_ptr_t dotu_dispatch_vector[dpnp_td_ns::num_types];

PYBIND11_MODULE(_blas_impl, m)
{
init_dispatch_vectors_tables();

using arrayT = dpctl::tensor::usm_ndarray;
using arrayT = dpnp::tensor::usm_ndarray;
using event_vecT = std::vector<sycl::event>;

{
Expand Down
2 changes: 1 addition & 1 deletion dpnp/backend/extensions/blas/dot.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@
namespace dpnp::extensions::blas
{
namespace mkl_blas = oneapi::mkl::blas;
namespace type_utils = dpctl::tensor::type_utils;
namespace type_utils = dpnp::tensor::type_utils;

template <typename T>
static sycl::event dot_impl(sycl::queue &exec_q,
Expand Down
24 changes: 12 additions & 12 deletions dpnp/backend/extensions/blas/dot_common.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@

#include <pybind11/pybind11.h>

// dpctl tensor headers
// dpnp tensor headers
Comment thread
ndgrigorian marked this conversation as resolved.
#include "utils/memory_overlap.hpp"
#include "utils/output_validation.hpp"
#include "utils/type_dispatch.hpp"
Expand All @@ -51,14 +51,14 @@ typedef sycl::event (*dot_impl_fn_ptr_t)(sycl::queue &,
char *,
const std::vector<sycl::event> &);

namespace dpctl_td_ns = dpctl::tensor::type_dispatch;
namespace dpnp_td_ns = dpnp::tensor::type_dispatch;
namespace py = pybind11;

std::pair<sycl::event, sycl::event>
dot_func(sycl::queue &exec_q,
const dpctl::tensor::usm_ndarray &vectorX,
const dpctl::tensor::usm_ndarray &vectorY,
const dpctl::tensor::usm_ndarray &result,
const dpnp::tensor::usm_ndarray &vectorX,
const dpnp::tensor::usm_ndarray &vectorY,
const dpnp::tensor::usm_ndarray &result,
const std::vector<sycl::event> &depends,
const dot_impl_fn_ptr_t *dot_dispatch_vector)
{
Expand All @@ -84,7 +84,7 @@ std::pair<sycl::event, sycl::event>
", but a 0-dimensional array is expected.");
}

auto const &overlap = dpctl::tensor::overlap::MemoryOverlap();
auto const &overlap = dpnp::tensor::overlap::MemoryOverlap();
if (overlap(vectorX, result)) {
throw py::value_error(
"The first input array and output array are overlapping "
Expand All @@ -96,17 +96,17 @@ std::pair<sycl::event, sycl::event>
"segments of memory");
}

if (!dpctl::utils::queues_are_compatible(
if (!dpnp::utils::queues_are_compatible(
exec_q,
{vectorX.get_queue(), vectorY.get_queue(), result.get_queue()})) {
throw py::value_error(
"USM allocations are not compatible with the execution queue.");
}

const int src_nelems = 1;
dpctl::tensor::validation::CheckWritable::throw_if_not_writable(result);
dpctl::tensor::validation::AmpleMemory::throw_if_not_ample(result,
src_nelems);
dpnp::tensor::validation::CheckWritable::throw_if_not_writable(result);
dpnp::tensor::validation::AmpleMemory::throw_if_not_ample(result,
src_nelems);

const py::ssize_t x_size = vectorX.get_size();
const py::ssize_t y_size = vectorY.get_size();
Expand All @@ -125,7 +125,7 @@ std::pair<sycl::event, sycl::event>
throw py::value_error("Given arrays must be of the same type.");
}

auto array_types = dpctl_td_ns::usm_ndarray_types();
auto array_types = dpnp_td_ns::usm_ndarray_types();
const int type_id = array_types.typenum_to_lookup_id(vectorX_typenum);

dot_impl_fn_ptr_t dot_fn = dot_dispatch_vector[type_id];
Expand Down Expand Up @@ -163,7 +163,7 @@ std::pair<sycl::event, sycl::event>
sycl::event dot_ev = dot_fn(exec_q, n, x_typeless_ptr, incx, y_typeless_ptr,
incy, r_typeless_ptr, depends);

sycl::event args_ev = dpctl::utils::keep_args_alive(
sycl::event args_ev = dpnp::utils::keep_args_alive(
exec_q, {vectorX, vectorY, result}, {dot_ev});

return std::make_pair(args_ev, dot_ev);
Expand Down
2 changes: 1 addition & 1 deletion dpnp/backend/extensions/blas/dotc.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@
namespace dpnp::extensions::blas
{
namespace mkl_blas = oneapi::mkl::blas;
namespace type_utils = dpctl::tensor::type_utils;
namespace type_utils = dpnp::tensor::type_utils;

template <typename T>
static sycl::event dotc_impl(sycl::queue &exec_q,
Expand Down
2 changes: 1 addition & 1 deletion dpnp/backend/extensions/blas/dotu.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@
namespace dpnp::extensions::blas
{
namespace mkl_blas = oneapi::mkl::blas;
namespace type_utils = dpctl::tensor::type_utils;
namespace type_utils = dpnp::tensor::type_utils;

template <typename T>
static sycl::event dotu_impl(sycl::queue &exec_q,
Expand Down
28 changes: 14 additions & 14 deletions dpnp/backend/extensions/blas/gemm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@
// utils extension header
#include "ext/common.hpp"

// dpctl tensor headers
// dpnp tensor headers
#include "utils/memory_overlap.hpp"
#include "utils/output_validation.hpp"
#include "utils/type_utils.hpp"
Expand All @@ -45,7 +45,7 @@ namespace dpnp::extensions::blas
{
namespace mkl_blas = oneapi::mkl::blas;
namespace py = pybind11;
namespace type_utils = dpctl::tensor::type_utils;
namespace type_utils = dpnp::tensor::type_utils;

using ext::common::init_dispatch_table;

Expand All @@ -64,8 +64,8 @@ typedef sycl::event (*gemm_impl_fn_ptr_t)(sycl::queue &,
const bool,
const std::vector<sycl::event> &);

static gemm_impl_fn_ptr_t gemm_dispatch_table[dpctl_td_ns::num_types]
[dpctl_td_ns::num_types];
static gemm_impl_fn_ptr_t gemm_dispatch_table[dpnp_td_ns::num_types]
[dpnp_td_ns::num_types];

template <typename Tab, typename Tc>
static sycl::event gemm_impl(sycl::queue &exec_q,
Expand Down Expand Up @@ -153,9 +153,9 @@ static sycl::event gemm_impl(sycl::queue &exec_q,

std::tuple<sycl::event, sycl::event, bool>
gemm(sycl::queue &exec_q,
const dpctl::tensor::usm_ndarray &matrixA,
const dpctl::tensor::usm_ndarray &matrixB,
const dpctl::tensor::usm_ndarray &resultC,
const dpnp::tensor::usm_ndarray &matrixA,
const dpnp::tensor::usm_ndarray &matrixB,
const dpnp::tensor::usm_ndarray &resultC,
const std::vector<sycl::event> &depends)
{
const int matrixA_nd = matrixA.get_ndim();
Expand All @@ -167,7 +167,7 @@ std::tuple<sycl::event, sycl::event, bool>
"Input and output matrices must be two-dimensional.");
}

auto const &overlap = dpctl::tensor::overlap::MemoryOverlap();
auto const &overlap = dpnp::tensor::overlap::MemoryOverlap();
if (overlap(matrixA, resultC)) {
throw py::value_error(
"The first input array and output array are overlapping "
Expand All @@ -179,7 +179,7 @@ std::tuple<sycl::event, sycl::event, bool>
"segments of memory");
}

if (!dpctl::utils::queues_are_compatible(
if (!dpnp::utils::queues_are_compatible(
exec_q,
{matrixA.get_queue(), matrixB.get_queue(), resultC.get_queue()})) {
throw py::value_error(
Expand All @@ -206,9 +206,9 @@ std::tuple<sycl::event, sycl::event, bool>
}

const std::size_t src_nelems = m * n;
dpctl::tensor::validation::CheckWritable::throw_if_not_writable(resultC);
dpctl::tensor::validation::AmpleMemory::throw_if_not_ample(resultC,
src_nelems);
dpnp::tensor::validation::CheckWritable::throw_if_not_writable(resultC);
dpnp::tensor::validation::AmpleMemory::throw_if_not_ample(resultC,
src_nelems);

const bool is_matrixA_f_contig = matrixA.is_f_contiguous();
const bool is_matrixB_f_contig = matrixB.is_f_contiguous();
Expand Down Expand Up @@ -300,7 +300,7 @@ std::tuple<sycl::event, sycl::event, bool>
throw py::value_error("matrixA and matrixB must be of the same type.");
}

auto array_types = dpctl_td_ns::usm_ndarray_types();
auto array_types = dpnp_td_ns::usm_ndarray_types();
const int matrixAB_type_id =
array_types.typenum_to_lookup_id(matrixA_typenum);
const int resultC_type_id =
Expand All @@ -322,7 +322,7 @@ std::tuple<sycl::event, sycl::event, bool>
a_typeless_ptr, lda, b_typeless_ptr, ldb,
r_typeless_ptr, ldc, is_row_major, depends);

sycl::event args_ev = dpctl::utils::keep_args_alive(
sycl::event args_ev = dpnp::utils::keep_args_alive(
exec_q, {matrixA, matrixB, resultC}, {gemm_ev});

return std::make_tuple(args_ev, gemm_ev, is_row_major);
Expand Down
12 changes: 6 additions & 6 deletions dpnp/backend/extensions/blas/gemm.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,16 +37,16 @@ namespace dpnp::extensions::blas
{
extern std::tuple<sycl::event, sycl::event, bool>
gemm(sycl::queue &exec_q,
const dpctl::tensor::usm_ndarray &matrixA,
const dpctl::tensor::usm_ndarray &matrixB,
const dpctl::tensor::usm_ndarray &resultC,
const dpnp::tensor::usm_ndarray &matrixA,
const dpnp::tensor::usm_ndarray &matrixB,
const dpnp::tensor::usm_ndarray &resultC,
const std::vector<sycl::event> &depends);

extern std::tuple<sycl::event, sycl::event, bool>
gemm_batch(sycl::queue &exec_q,
const dpctl::tensor::usm_ndarray &matrixA,
const dpctl::tensor::usm_ndarray &matrixB,
const dpctl::tensor::usm_ndarray &resultC,
const dpnp::tensor::usm_ndarray &matrixA,
const dpnp::tensor::usm_ndarray &matrixB,
const dpnp::tensor::usm_ndarray &resultC,
const std::vector<sycl::event> &depends);

extern void init_gemm_dispatch_table(void);
Expand Down
26 changes: 13 additions & 13 deletions dpnp/backend/extensions/blas/gemm_batch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@
// utils extension header
#include "ext/common.hpp"

// dpctl tensor headers
// dpnp tensor headers
#include "utils/memory_overlap.hpp"
#include "utils/output_validation.hpp"
#include "utils/type_utils.hpp"
Expand All @@ -45,7 +45,7 @@ namespace dpnp::extensions::blas
{
namespace mkl_blas = oneapi::mkl::blas;
namespace py = pybind11;
namespace type_utils = dpctl::tensor::type_utils;
namespace type_utils = dpnp::tensor::type_utils;

using ext::common::init_dispatch_table;

Expand All @@ -70,7 +70,7 @@ typedef sycl::event (*gemm_batch_impl_fn_ptr_t)(
const std::vector<sycl::event> &);

static gemm_batch_impl_fn_ptr_t
gemm_batch_dispatch_table[dpctl_td_ns::num_types][dpctl_td_ns::num_types];
gemm_batch_dispatch_table[dpnp_td_ns::num_types][dpnp_td_ns::num_types];

template <typename Tab, typename Tc>
static sycl::event gemm_batch_impl(sycl::queue &exec_q,
Expand Down Expand Up @@ -212,9 +212,9 @@ void standardize_strides_to_zero(std::vector<py::ssize_t> &strides,

std::tuple<sycl::event, sycl::event, bool>
gemm_batch(sycl::queue &exec_q,
const dpctl::tensor::usm_ndarray &matrixA,
const dpctl::tensor::usm_ndarray &matrixB,
const dpctl::tensor::usm_ndarray &resultC,
const dpnp::tensor::usm_ndarray &matrixA,
const dpnp::tensor::usm_ndarray &matrixB,
const dpnp::tensor::usm_ndarray &resultC,
const std::vector<sycl::event> &depends = {})
{
const int matrixA_nd = matrixA.get_ndim();
Expand All @@ -225,7 +225,7 @@ std::tuple<sycl::event, sycl::event, bool>
throw py::value_error("The given arrays have incorrect dimensions.");
}

auto const &overlap = dpctl::tensor::overlap::MemoryOverlap();
auto const &overlap = dpnp::tensor::overlap::MemoryOverlap();
if (overlap(matrixA, resultC)) {
throw py::value_error("Input array 1 and output array are overlapping "
"segments of memory");
Expand All @@ -235,7 +235,7 @@ std::tuple<sycl::event, sycl::event, bool>
"segments of memory");
}

if (!dpctl::utils::queues_are_compatible(
if (!dpnp::utils::queues_are_compatible(
exec_q,
{matrixA.get_queue(), matrixB.get_queue(), resultC.get_queue()})) {
throw py::value_error(
Expand All @@ -262,9 +262,9 @@ std::tuple<sycl::event, sycl::event, bool>
"the number of columns in result array.");
}
const std::int64_t src_nelems = batch_size * m * n;
dpctl::tensor::validation::CheckWritable::throw_if_not_writable(resultC);
dpctl::tensor::validation::AmpleMemory::throw_if_not_ample(resultC,
src_nelems);
dpnp::tensor::validation::CheckWritable::throw_if_not_writable(resultC);
dpnp::tensor::validation::AmpleMemory::throw_if_not_ample(resultC,
src_nelems);

std::vector<py::ssize_t> a_stride = matrixA.get_strides_vector();
std::vector<py::ssize_t> b_stride = matrixB.get_strides_vector();
Expand Down Expand Up @@ -374,7 +374,7 @@ std::tuple<sycl::event, sycl::event, bool>
throw py::value_error("matrixA and matrixB must be of the same type.");
}

auto array_types = dpctl_td_ns::usm_ndarray_types();
auto array_types = dpnp_td_ns::usm_ndarray_types();
const int matrixAB_type_id =
array_types.typenum_to_lookup_id(matrixA_typenum);
const int resultC_type_id =
Expand All @@ -397,7 +397,7 @@ std::tuple<sycl::event, sycl::event, bool>
strideb, stridec, transA, transB, a_typeless_ptr,
b_typeless_ptr, r_typeless_ptr, is_row_major, depends);

sycl::event args_ev = dpctl::utils::keep_args_alive(
sycl::event args_ev = dpnp::utils::keep_args_alive(
exec_q, {matrixA, matrixB, resultC}, {gemm_batch_ev});

return std::make_tuple(args_ev, gemm_batch_ev, is_row_major);
Expand Down
Loading
Loading